I dont think LLMs will or should replace properly written dialogue.
Where they would shine is just generating inane background chatter. So instead of hearing an NPC say “i took an arrow to the knee”, or “jesus christ be praised! Henry’s come to see us!” 300 times, an LLM could generate some short one liners that are a bit more dynamic. It would go a long way to making the world feel more alive.
I bet that it’s not too expensive to write a lot of human-written dialog, but that hiring a bunch of voice actors to act out minor lines – especially if a given character has only a few lines – it is probably the more-expensive bit. Like, I think that a human dialog writer could probably affordably put together enough dialog that a player wouldn’t really exhaust it, but that you’d want to make any synthesis of the lines not have a lot of extra cost.
New text to speech models are incredible these days…
Again, we shouldnt replace actual voice actors for main dialog. But for generating thousands of lines of background chatter (which nobody would have time or resources to make anyway) LLM writing paired with text to speech could really help flesh out a living game world
I dont think LLMs will or should replace properly written dialogue.
Where they would shine is just generating inane background chatter. So instead of hearing an NPC say “i took an arrow to the knee”, or “jesus christ be praised! Henry’s come to see us!” 300 times, an LLM could generate some short one liners that are a bit more dynamic. It would go a long way to making the world feel more alive.
That’s a thought.
considers
I still think that the limiting factor there is more one of speech synth than writing dialog. Like, “arrow to the knee” is Skyrim, right?
kagis
Yeah. And those were voiced.
Similarly, you had Fallout: New Vegas with stuff like “patrolling the Mojave almost makes you wish for a nuclear winter”.
I bet that it’s not too expensive to write a lot of human-written dialog, but that hiring a bunch of voice actors to act out minor lines – especially if a given character has only a few lines – it is probably the more-expensive bit. Like, I think that a human dialog writer could probably affordably put together enough dialog that a player wouldn’t really exhaust it, but that you’d want to make any synthesis of the lines not have a lot of extra cost.
New text to speech models are incredible these days…
Again, we shouldnt replace actual voice actors for main dialog. But for generating thousands of lines of background chatter (which nobody would have time or resources to make anyway) LLM writing paired with text to speech could really help flesh out a living game world